Designing Next Generation Exhaust Aftertreatment Systems Using Machine Learning

With ever tightening emission standards, the automotive industry is continuously seeking novel ways to improve the aftertreatment system (ATS). Exhaust treatment systems using diesel emission fluid (DEF), in conjunction with selective catalytic reduction (SCR) and diesel oxidation converters (DOC), have been gaining popularity in the heavy equipment industry. Spraying DEF (mixture of urea and water) into the exhaust flow can convert harmful NOx gases into N₂ and H₂O. Design of ATSs focuses on high evaporation rate and uniform mixing of ammonia at the entrance to the SCR catalyst. This study applied support vector regressor (SVR), a machine learning (ML) method to a database of computational fluid dynamics (CFD) simulations to develop a highly efficient mixer with high heat exchange characteristics. Over 500 mixer designs were evaluated using CFD and were then used to train the SVR model. The trained ML model was then used as a surrogate to the CFD and coupled with the genetic algorithm (GA), an optimization technique, to further refine the design parameters. The optimal design obtained from this methodology showed a remarkable performance improvement compared to the baseline.